Imagine if your next experiment knew the outcome of your last thousand...

Imagine if your next experiment knew the outcome of your last thousand.

At ReactWise, we’ve spent the last year focused on two things: getting the highest quality data - and enabling our algorithms to make the best use of any prior data.

In our high-throughput experimentation lab, we’ve generated thousands of high-quality datapoints across different chemistries, catalysts, and process conditions. That data became the foundation to fuel MemoryBO, our proprietary multi-task Bayesian optimization algorithm.

Traditional Bayesian optimization starts every campaign from scratch. MemoryBO doesn’t. 

It learns from past experiments, transferring knowledge between related systems and intelligently warm-starting new optimizations. 

The result is faster convergence, smarter exploration, and identification of high-yield regions that standard BO would take much longer to find.

The plot below shows it in action during optimization of an amide coupling reaction - a clear performance boost over traditional BO, achieving results that would otherwise remain undiscovered. The Y-axis represents regret - the distance to the best achievable value, approximated by the top experimental outcome.

We’re pushing towards a future where every experiment counts twice: once for its immediate outcome, and once for what it teaches the next optimization.

This is how chemistry learns from itself.

‍